MATS mentors should really reconsider requiring (or putting any value on the results of) the "AI Assisted Coding Assessment". CodeSignal's "AI coding assistant" is a 2024-era chat bot (GPT-4o) and a text editor. Using this efficiently is a very different skill from both pre-AI coding and using modern reasoning models[1] with standard tooling[2].
The first net-helpful coding agents came out sometime in 2025, and the earliest that can be used anything remotely like modern models was Opus 4.5 in November of that year. GPT-4o isn't even a reasoning model.
It's not gpt-4o – I just did the assessment and the chatbot claimed (incorrectly, presumably) to be Claude 3.5 Sonnet with a knowledge cutoff in early 2024 (I think it was probably actually Sonnet 4.5?). I found it really annoying to deal with (e.g. I had to keep copypasting code from the chatbot to the editor, it talked in an annoying way (e.g. lots of emojis), and made a lot mistakes), and probably would have done better if I just hadn't used it.
I think Fable 5.1 with a proper harness would've probably have been able to one-shot the task.
4Brendan Long
All of the documentation I've seen says it's GPT-4o, so that's even more confusing.
I had the same problems, and it also kept re-introducing the same bug on every edit. If I had realized what I was getting into I would have used it in a very different way, but there was no way to practice either.